Abstract: Aiming at the problem of wood defect image perception and quality decision-making in typical bionic intelligent algorithm, the defect image distortion was serious, the variance of prior feature extraction of defect image fluctuates frequently, the gray level segmentation of defect image with uneven texture is invalid, the generalization ability and learning ability of different wood texture are unbalanced, and the optimal convergence speed is delayed with the defect dimension, a model of wood defect image reconstruction and quality evaluation based on deep reinforcement learning was proposed.By introducing the deep learning mechanism and using the deep residual network for iterative training, we can realize the real-time and efficient reconstruction of the multi-dimensional defect image of different wood, build a panoramic autonomous perception model for the fine segmentation and feature extraction of multi-dimensional defect of different wood, and build a large data level shared resource pool of wood defect features;By introducing reinforcement learning mechanism and using depth deterministic strategy gradient algorithm, a high-dimensional decision mapping among iterative updating of defect features, independent decision-making, panoramic visibility, depth prediction and wood quality evaluation was established, which realized the horizontal sharing integration of multi-dimensional difference wood defect image reconstruction and quality evaluation. Taking an economic forest in Nanhu forest farm area of Huzhou city, Zhejiang province as the evaluation carrier, the engineering application analysis of the model was carried out. The verification results showed that the model proposed in this paper can better realize the multi-dimensional defect perception and reconstruction of wood, the autonomous intelligent decision-making of global optimal quality evaluation, and has the obvious ability of sensing autonomy, reconstruction reproducibility, autonomous decision-making, model generalization, etc show superiority.

Key words: wood defect detection, image reconstruction, deep reinforcement learning, quality evaluation, self perception and decision making